1 / 77

Advanced Methods and Analysis for the Learning and Social Sciences

Advanced Methods and Analysis for the Learning and Social Sciences. PSY505 Spring term, 2012 February 6, 2012. Today’s Class. Advanced BKT and Learning Decomposition. Learning Decomposition (Beck et al., 2006). A, -b, B: Free parameters t1, t2: Number of practice of types 1, 2

sancho
Download Presentation

Advanced Methods and Analysis for the Learning and Social Sciences

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Advanced Methods and Analysis for the Learning and Social Sciences PSY505Spring term, 2012 February 6, 2012

  2. Today’s Class • Advanced BKT and Learning Decomposition

  3. Learning Decomposition(Beck et al., 2006) • A, -b, B: Free parameters • t1, t2: Number of practice of types 1, 2 • For example, with help request, without help request

  4. Interpretation • A: Performance on first trial • b: Improvement per trial • B: Practice type weight • If B>1, t1 leads to more learning than t2 • If B<1, t2 leads to more learning than t1 • If B=1, t1 and t2 are equal

  5. Example • Let’s make up some data where students improve a little with t1 and a lot with t2 • Then we’ll see how LD fits it

  6. Advantages?

  7. Advantages? • Very easy to interpret

  8. Disadvantages?

  9. Disadvantages? • Statistical significance testing is hard • Requires computing confidence intervals for a parameter in exponential regression, and determining if 1 is inside confidence interval

  10. Variant • Take the log • Log Performance = -bBt1 - bt2 + c • Conduct logistic regression • Log performance = dt1 + et2 + c • Possible to compute standard error for d and e • Therefore, possible to compute a statistical test as to whether d and e are significantly different

  11. Comments? Questions?

  12. Advanced BKT

  13. Classic BKT p(T) Not learned Learned p(L0) p(G) 1-p(S) correct correct Two Learning Parameters p(L0) Probability the skill is already known before the first opportunity to use the skill in problem solving. p(T) Probability the skill will be learned at each opportunity to use the skill. Two Performance Parameters p(G) Probability the student will guess correctly if the skill is not known. p(S) Probability the student will slip (make a mistake) if the skill is known.

  14. Extensions to BKT • Largely take the form of relaxing the assumption that parameters vary by skill, but are constant for all other factors

  15. Beck et al.’s (2008) Help Model p(T|H) Not learned Learned p(L0|H), p(L0|~H) p(T|~H) 1-p(S|~H) 1-p(S|H) p(G|~H), p(G|H) correct correct

  16. Fit… • The same way as regular BKT

  17. Beck et al.’s (2008) Help Model • Parameters per skill: 8 • Effect on future prediction: slightly worse

  18. Beck et al.’s (2008) Help Model • What do these parameters suggest?

  19. Beck et al.’s (2008) Help Model • Learning decomposition disagrees; which should we trust?

  20. Moment-By-Moment Learning Model(Baker, Goldstein, & Heffernan, 2010) Probability you Just Learned p(J) p(T) Not learned Learned p(L0) p(G) 1-p(S) correct correct

  21. P(J) • P(T) = chance you will learn if you didn’t know it • P(J) = probability you JustLearned • P(J) = P(~Ln ^ T)

  22. P(J) is distinctfrom P(T) • P(T) is held fixed for each skill • P(J) is contextual (calculated for each problem) • You *could* make P(T) contextual too • But a contextual P(T) would not be interpretable for high P(Ln) • For example: • P(Ln) = 0.1 • P(T) = 0.6 • P(J) = 0.54 • P(Ln) = 0.96 • P(T) = 0.6 • P(J) = 0.02 • Learning! • Little Learning

  23. Labeling P(J) • Based on this concept: • “The probability a student did not know a skill but then learns it by doing the current problem, given their performance on the next two.” P(J) = P(~Ln ^ T | A+1+2 ) *For full list of equations, see Baker, Goldstein, & Heffernan (2010)

  24. Breaking down P(~Ln ^ T | A+1+2 ) • We can calculate P(~Ln ^ T | A+1+2 ) with an application of Bayes’ theorem • P(~Ln ^ T | A+1+2 ) = Bayes’ Theorem: P(A | B) = P(A+1+2 | ~Ln ^ T) * P(~Ln ^ T) P (A+1+2 ) P(B | A) * P(A) P(B)

  25. Breaking down P(A+1+2 ) • P(~Ln ^ T ) is computed with BKT building blocks {P(~Ln), P(T)} • P(A+1+2 )is a function of the only three relevant scenarios, {Ln, ~Ln ^ T, ~Ln ^ ~T}, and their contingent probabilities • P(A+1+2 ) = P(A+1+2 | Ln) P(Ln) + P(A+1+2 | ~Ln ^ T) P(~Ln ^ T) + P(A+1+2 | ~Ln ^ ~T) P(~Ln ^ ~T)

  26. Breaking down P(A+1+2 | Ln) P(Ln):One Example • P(A+1+2 = C, C | Ln) = P(~S)P(~S) • P(A+1+2 = C, ~C | Ln) = P(~S)P(S) • P(A+1+2 = ~C, C | Ln) = P(S)P(~S) • P(A+1+2 = ~C, ~C | Ln) = P(S)P(S) • (Correct marked C, wrong marked ~C)

  27. Features of P(J) • Distilled from logs of student interactions with tutor software • Broadly capture behavior indicative of learning • Selected from same initial set of features previously used in detectors of • gaming the system (Baker, Corbett, Roll, & Koedinger, 2008) • off-task behavior (Baker, 2007) • carelessness (Baker, Corbett, & Aleven, 2008)

  28. Features of P(J) • All features use only first response data • Later extension to include subsequent responses only increased model correlation very slightly – not significantly(Baker, Goldstein, & Heffernan, 2011)

  29. Features of P(J) • Argument could be made that using BKT probabilities (Ln) in the prediction of the label (~Ln ^ T) is wrong • We consider this to be valid – theoretically important part of model is the T, not the Ln • The model maintains a 0.301 correlation coefficient even without Ln or Ln-1

  30. The final model

  31. Interpretation • P(J) is higher following incorrect responses • People learn from their mistakes • However, P(J) decreases as the total number of times student got this skill wrong increases • Might need intervention not available in the tutor

  32. Interpretation • P(J) is lower following help requests • P(J) is higher when help has been used recently, i.e. in the last 5 and/or 8 steps

  33. Replication • We replicated this result in another intelligent tutor, ASSISTments (Razzaq et al., 2007) • Correlation between models and labels: 0.397

  34. Model use • Does learning in intelligent tutors have more of a character of • gradual learning (such as strengthening of a memory association – cf. Newell & Rosenbloom, 1981; Heathcote, Brown, & Mewhort, 2000) • or learning given to “eureka”/insight moments, where a skill is understood suddenly? (Lindstrom & Gulz, 2008) • Does this vary by skill?

  35. Model use • Both types of learning known to occur but the conditions leading to each are still incompletely known (Bowden et al., 2005) • Opportunity to study these issues in real learning, which is still uncommon – most research in insight takes place for toy problems in lab settings (as pointed out by Bowden et al., 2005; Kounios et al., 2008)

  36. To investigate this • We can plot P(J) over time, and see how “spiky” the graph is • Spikes indicating moments of higher learning

  37. Real Data for A Student “Entering a common multiple” P(J) OPTOPRAC

  38. Real Data for Same Student “Identifying the converted value in the problem statement of a scaling problem” P(J) OPTOPRAC

  39. As you can see… • One skill was learned gradually, the other skill was learned suddenly • Note that the first graph had *two* spikes • This was actually very common in the data, even more common than single spikes • We are still investigating why this happens

  40. We can quantify the difference between these graphs • We can quantify the degree to which a learning sequence involves a “eureka” moment, through a metric we call “spikiness” • For a given student/skill pair, spikiness = • Max P(J)/Avg P(J) • Scaled from 1 to infinity

  41. Looking at spikiness • We only consider action sequences at least 6 problem steps long • (Shorter sequences tend to more often look spiky, which is a mathematical feature of using a within-sequence average) • We only consider the first 20 problem steps • After that, the student is probably floundering

  42. Spikiness by skill • Min: 1.12 • Max: 113.52 • Avg: 8.55 • SD: 14.62 • Future work: What characterizes spiky skills versus gradually-learned skills?

  43. Spikiness by student • Min: 2.22 • Max: 21.81 • Avg: 6.81 • SD: 3.09 • Students are less spiky than skills

  44. Interestingly • The correlation between a student’s spikiness, and their final average P(Ln) across skills is a high 0.71, statistically significantly different than chance • Suggests that learning spikes may be an early predictor of whether a student is going to achieve good learning of specific material • May someday be the basis of better knowledge tracing

  45. Also… • The cross-validated correlation between a student’s average P(J) • And their performance on a test of preparation for future learning (cf. Bransford & Schwartz, 1999) where the student needs to read a text to learn a related skill • Is 0.35 • statistically significantly different than chance • higher than the predictive power of Bayesian Knowledge-Tracing for this same test(Baker, Gowda, & Corbett, 2011) • Suggests that this model is capturing deep aspects of learning not captured in existing knowledge assessments

  46. Worth Noting • Generally, across all actions on a skill, the P(J) model doesn’t quite add up to a total of 1 • In general, our model is representative of P(J) at lower levels but tends to underestimate the height of spikes • May be a result of using a linear modeling approach for a fundamentally non-linear phenomenon • May also be that P(J) is actually too high in the training labels (where it often sums up to significantly more than 1) • Could be normalized -- for the purposes of spikiness analyses, we believe the model biases towards seeing less total spikiness

  47. Baker, Corbett, & Aleven’s (2008)Contextual Guess and Slip model p(T) Not learned Learned p(L0) p(G) 1-p(S) correct correct

  48. Contextual Slip: The Big Idea • Why one parameter for slip • For all situations • For each skill • When we can have a different prediction for slip • For each situation • Across all skills

  49. In other words • P(S) varies according to context • For example • Perhaps very quick actions are more likely to be slips • Perhaps errors on actions which you’ve gotten right several times in a row are more likely to be slips

  50. Baker, Corbett, & Aleven’s (2008)Contextual Guess and Slip model • Guess and slip fit using contextual models across all skills • Parameters per skill: 2 + (P (S) model size)/skills + (P (G) model size)/skills • 2.11 parameters per skill

More Related